PORCA: Modeling and Planning for Autonomous Driving Among Many Pedestrians

Luo, Yuanfu; Cai, Panpan; Bera, Aniket; Hsu, David; Lee, Wee Sun; Manocha, Dinesh · 2018 · OpenAlex-citations

DOI: 10.1109/lra.2018.2852793

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of autonomous vehicle navigation in densely populated environments, such as shopping malls or hospital complexes, where vehicles must drive safely and smoothly among many pedestrians. The core problem is the uncertainty inherent in pedestrian behavior, which is driven by both global navigation intentions (e.g., final destinations) and local interactions with other agents. Existing methods often fail to account for these uncertainties or rely on simplified motion models that do not generalize well. The authors propose a planning system that integrates a novel pedestrian motion model, Pedestrian ORCA (PORCA), with a Partially Observable Markov Decision Process (POMDP) to handle uncertainty in real time. The methodology centers on PORCA, an extension of the Optimal Reciprocal Collision Avoidance (ORCA) algorithm. Standard ORCA suffers from the "freezing pedestrian" problem, where agents slow down excessively or stop when encountering head-on conflicts, and it assumes all agents are holonomic, ignoring the non-holonomic constraints of vehicles. PORCA addresses these limitations by modifying the objective function to include a "patience" term that encourages agents to maintain speed and explore detours rather than stopping. Additionally, PORCA assigns greater collision avoidance responsibility to pedestrians when interacting with vehicles, reflecting the vehicle’s limited maneuverability. This motion model is embedded within a POMDP framework, where pedestrian intentions are treated as hidden states. The system uses a belief tracker to maintain probability distributions over these intentions and a parallel DESPOT algorithm to solve the POMDP efficiently, planning optimal vehicle speed control while a separate Hybrid A* planner handles steering. Experiments conducted on a robot scooter in a campus plaza with a pedestrian density of nearly one person per square meter demonstrate the effectiveness of the approach. The results show that PORCA predicts pedestrian motions more accurately than prior models. By integrating this model with POMDP planning, the autonomous vehicle successfully avoids collisions while reaching its goals more efficiently and smoothly compared to methods that ignore intentions or interactions. The system operates in real time, re-planning steering and speed at 3 Hz. The significance of this work lies in its principled handling of uncertainty in social navigation. By explicitly modeling both pedestrian intentions and local interactions within a probabilistic framework, the system enables robust autonomous driving in complex, crowded scenarios. This approach overcomes the limitations of static obstacle avoidance or independent motion assumptions, providing a scalable solution for autonomous vehicles operating in public spaces where human-robot interaction is frequent and unpredictable.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• anticipation